How AI search ranking actually works
Plain-language answers to what every AEO metric means and, roughly, how it's calculated — the same reference our customers use to read their own dashboard, open to anyone asking how this works.
Want to see where your own site stands first?
Run a free reviewThe basics
What's a “check”?
One AI platform's answer to one of your prompts, on one scan. If you track 8 prompts across 5 platforms, a single scan produces 8 × 5 = 40 checks for your site — plus the same again for each competitor you track, since we ask about them from the same answer.
What counts as a prompt “mentioning” me?
Claude reads the AI platform's raw answer and judges whether your site, brand, or URL was named anywhere in it — by name, not just implied. A mention doesn't require a link or a recommendation, just that you were named as one of the options.
What's “visibility %” / “mention rate”?
The share of checks where you were mentioned, out of all checks in the window shown. 60% visibility over the last 30 days means you showed up in 6 out of every 10 platform/prompt checks in that period.
Do you track Google AI Overviews?
Not directly — we check Claude, ChatGPT, Grok, Perplexity, and Gemini's chat answers, not Google's AI Overviews search feature specifically. The two overlap in practice (Overviews are powered by Gemini), and the same fundamentals apply: if the AI can't tell what you offer, it can't mention you in either place. Direct Overviews tracking is a fair ask — tell us if it'd change your plan.
Rank & position
What does the position number (#1, #2, …) mean?
Your rank among every distinct product, company, or tool the AI's answer named — not just among the competitors you're tracking. If an answer lists five options and you're the second one named, that's #2, even if none of your tracked competitors appear in that particular answer.
Why don't I have a position on every check?
Position only applies when you were mentioned. “—” means you weren't named in that answer at all, so there's no rank to show.
Sentiment
How is sentiment decided?
For every mention, Claude judges how you were framed in that specific answer: positive (recommended, praised, framed as a strong fit), neutral (just listed or named with no clear framing either way), or negative (criticized or framed as a poor fit). It's a judgment call on wording, not a keyword match — the same as reading the answer yourself and asking “would I feel good about this?”
Why do I care about sentiment if I'm already mentioned?
Being named isn't the same as being recommended. A model that lists you fourth with “also worth a look” is a very different outcome than one that opens with you as the clear pick — visibility % alone can't tell those apart, sentiment can.
MyAEO Rank
What is MyAEO Rank?
A single 0-100 score blending three things visibility % alone doesn't capture: whether you're mentioned at all (presence), how prominently when you are (rank quality), and whether the mention reads well (sentiment quality). It's our own composite, not an industry-standard metric — shown as a breakdown (Presence / Rank / Sentiment, in the stat cards) so you can see what's driving the number, not just the number itself.
How exactly is it calculated?
Each of the three components is scored 0-1 and multiplied by a weight (50/25/25 by default): Presence is just your mention rate. Rank quality averages, across your mentions, a score that's 1.0 at position #1 and falls off toward 0 by position #5+. Sentiment quality averages 1.0 for positive mentions, 0.5 for neutral, 0 for negative. The three weighted pieces are added together and rounded — that's the score.
Can I change the weighting?
Yes — from Settings, if presence matters more to you than polish (or vice versa), adjust the three weights (they must sum to 100). The chart and score update to reflect your weighting, not just the default.
Citations
What's the difference between a mention and a citation?
A mention is your name appearing in the answer text. A citation is when the platform also lists your URL among its source links — a stronger signal, since it means the platform is pointing readers back to your site, not just naming you.
Why do some platforms show “—” for citations instead of 0%?
Perplexity and Gemini list source links alongside their answers, so we can check whether your URL is among them. Claude, ChatGPT, and Grok don't expose source links the same way today, so citation tracking isn't applicable there — shown as unset rather than a false “0%,” which would wrongly imply we checked and found none.
Prompts & priority
Where do auto-generated prompts come from?
When you add a site, we scrape the pages you selected and identify the keyword phrases an AI assistant would associate with your business, then derive a realistic natural-language question for each one — the kind of thing a potential customer would actually type, not the keyword itself with a question mark on it.
What does marking a prompt “priority” do?
It adds that prompt to the “Priority prompts over time” chart, tracked separately from your aggregate visibility — useful for watching the handful of questions you're actively trying to move the needle on, without them getting averaged into everything else.
How does importing a keyword list turn into prompts?
Each keyword you paste becomes one prompt via a rewrite that stays close to your original wording rather than paraphrasing freely — these are terms you already know matter, so we don't want to drift from them. The order you paste them in becomes their priority, since a keyword list is usually already roughly sorted by importance.
What's the prompt library?
A saved set of prompts, separate from any one site, for the ones you'd want to reuse — check "also save to your library" when adding a custom prompt, and it shows up as a one-click add on every other site you track. Handy if you're tracking several sites in the same category and want the same handful of questions checked on each.
Competitors & other mentions
How are competitors ranked against me?
The same prompt, platform, and answer are used to judge you and every tracked competitor at once — not separate questions asked separately — so the ranking reflects one real answer's ordering, not results stitched together from different queries.
What's the “Other names AI mentions” panel?
Every check also captures any other product, company, or tool the answer named that you're not explicitly tracking. It's a free signal for who else shows up in your category — worth knowing about before deciding whether to add one as a tracked competitor.
What's “Share of voice”?
Your slice of every mention that happened across you and your tracked competitors, in the same set of answers — a category-relative number. It's different from visibility %, which is measured against your total checks regardless of whether competitors were mentioned at all. Share of voice only appears once you're tracking at least one competitor, since there's nothing to share otherwise.
Recommendations
How are the Recommendations generated?
Once a week (or on demand via the refresh button), we summarize your last 14 days of scans — mention rates vs. the 14 days before, which competitors win which prompts, sentiment mix, citation rate, and recurring untracked mentions — and turn that summary into 3-5 suggestions. Every recommendation has to cite a specific number from your own data; we deliberately avoid generic AEO advice with nothing behind it.
Why does it only update weekly?
Meaningful movement in AI answers takes more than a day or two to show up reliably, and a weekly cadence keeps it aligned with how often most plans scan. Use “Refresh recommendations” any time you want an up-to-date read sooner.
AI crawlability
What does the AI crawlability check actually check?
Whether your site's robots.txt disallows known AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and others), and whether you publish an llms.txt. A blocked crawler can't read your pages firsthand when someone asks an AI assistant about you directly — llms.txt is optional, but some platforms use it to understand a site faster.
Does blocking a crawler affect my scan results?
Not directly — we use Firecrawl to read your site for keyword/prompt generation regardless. But if GPTBot or ClaudeBot is blocked in production, that platform's own training and browsing can't see your site the way it sees an unblocked competitor's, which can suppress mentions over time independent of anything we measure here.
Scan cadence & history
How often does my data update?
Depends on your plan — weekly, daily, or on-demand only, shown on your dashboard. Every plan includes one free scan when you add a site; after that, on-demand re-scans depend on your plan tier.
Can I see past scans, not just the latest?
Yes — the ranking comparison table has a date picker that steps back through every past scan, using the same history that feeds the trend charts, so nothing is summarized away.
Have a question this doesn't answer? Contact us and we'll add it here.